Floating polygon soup

Abstract

This paper presents a new representation called floating polygon soup for applications like 3DTV and FTV (Free Viewpoint Television). This representation is based on 3D polygons and takes as input MVD data. It extends the previously proposed polygon soup representation [1] which is appropriate for both compression, transmission and rendering stages. The floating polygon soup conserves these advantages while also taking into account misalignments at the view synthesis stage due to modeling errors. The idea for reducing these misalignments is to morph the 3D geometry depending on the current viewpoint. Results show that artifacts in virtual views are reduced and objective quality is increased.